ElevenLabs makes Music v2.5 its default song model and blocks downloads built on another artist's work
ElevenLabs has released Music v2.5, the newest version of the song-generation model behind its ElevenMusic service, and made it the default for prompted and reference-based generation while keeping Music v2 available. The company calls the songs fuller and more natural, a description resting on its own listening test. Users keep the rights to what they generate, imitating an existing musician is not allowed, and downloads of tracks built on another artist's songs are blocked. A recent Universal Music Group licensing deal leaves this model out.
Artificial Intelligence··Night
Music v2.5 becomes the default and reaches the API
ElevenLabs, which generates voice and music with AI, has released Music v2.5 for ElevenMusic, its song-generation service. The new version is now the default model for generation from a text prompt and from a reference track, while Music v2 stays available. The model can also be reached inside the company's ElevenCreative, Studio and Flows tools and through the ElevenLabs API, so developers can generate tracks inside their own products. According to the company, songs from the new version sound fuller and more natural; that description rests on a blind listening test in which listeners preferred the new version most of the time, and the test is the company's own measurement.[1], [2]
Users keep the rights while imitation and derivative downloads are blocked
Users keep the rights to the tracks they generate. Imitating an existing musician is not allowed, and downloads of tracks based on another artist's songs are blocked. Commercial use depends on the industry and the purpose; on the free plan it requires crediting ElevenMusic and remains subject to the service's restrictions.[1], [2]
The Universal Music Group deal stops short of Music v2.5
ElevenLabs recently signed a licensing deal with the music company Universal Music Group. According to the company, that deal covers future, separate products and leaves Music v2.5 outside it. ElevenLabs says its existing music models were trained on licensed stems and music, without detailing which data was used.[1]
